spring-cloud / spring-cloud/spring-cloud-function

KafkaNull payload bypasses MessageConverterHelper and is passed unconverted to a typed Consumer, causing ClassCastException

Aperta
#1,448 0 commenti 0 reazioni 0 assegnatari Vedi su GitHub

Nessuno ha ancora preso questa issue.

Lingua principale
Java
Stelle
1.1k
Fork
641
Merge medio
11h 2m
PR unite (30g)
8

Descrizione

Describe the bug

When a Kafka consumer function is bound to a concrete argument type (e.g. Consumer<MyDomainType>), and the broker delivers a record with a null value — whether from log-compaction tombstone semantics, or simply a producer publishing an empty/null payload — Spring Cloud Stream's Kafka binder wraps it as a Message whose payload is KafkaNull.INSTANCE. SimpleFunctionRegistry.FunctionInvocationWrapper#convertInputIfNecessary detects this and returns the unconverted, raw Message as the function's input argument, instead of attempting conversion or signalling a conversion failure:

https://github.com/spring-cloud/spring-cloud-function/blob/main/spring-cloud-function-context/src/main/java/org/springframework/cloud/function/context/catalog/SimpleFunctionRegistry.java

else if (input instanceof Message) {
    input = this.filterOutHeaders((Message) input);
    if (((Message) input).getPayload().getClass().getName().equals("org.springframework.kafka.support.KafkaNull")) {
        return input;   // bypasses every converter and the MessageConverterHelper hook entirely
    }
    ...

The generated bridge method for the functional interface then tries to cast that Message to the function's declared argument type, throwing a bare ClassCastException from SimpleFunctionRegistry.FunctionInvocationWrapper#invokeConsumer.

Critically, this happens without ever calling any registered MessageConverter or MessageConverterHelper, so the extension point added to resolve #1168 (letting an application signal "this should be a fatal conversion failure, not passed to my function") has no effect here — the KafkaNull branch returns early before that machinery is ever reached. Frameworks and applications built on top of this therefore see a plain ClassCastException with no way to distinguish "the framework couldn't convert this" from "a bug in my own consumer code."

This is closely related to #1168, which covers the case where a registered converter throws during conversion. This issue covers a narrower, still-open gap: the KafkaNull payload never reaches a converter at all, so #1168's fix doesn't help here.

Environment

  • spring-cloud-function-context 4.3.4
  • spring-cloud-stream / spring-cloud-stream-binder-kafka 4.3.3
  • spring-kafka 3.3.16
  • (the shortcut appears structurally unchanged across the versions we checked, so this likely affects a wide range)

To Reproduce

  1. Define a functional Kafka consumer bound to a concrete type, e.g. Consumer<Event<SomeDomainType>>.
  2. Publish a Kafka record with a null value to the bound topic.
  3. Observe: the consumer function is invoked with a raw Message object instead of the declared type, throwing ClassCastException (e.g. class org.springframework.messaging.support.GenericMessage cannot be cast to class com.example.SomeDomainType).

We can provide a minimal standalone Spring Boot + Testcontainers-Kafka reproduction project if useful.

Expected behavior

When the payload is KafkaNull and the target function's declared argument type is not itself Message/KafkaNull-compatible, the framework should route through the same MessageConverterHelper.shouldFailIfCantConvert mechanism used elsewhere — either raising a MessageConversionException or letting the registered helper decide — rather than unconditionally passing the raw Message through. A function genuinely declared to accept Message<?>/KafkaNull (a real null-value-aware consumer) should keep working exactly as it does today; only the mismatched-type case should change.

Suggested fix

In the KafkaNull branch of convertInputIfNecessary, check whether the function's declared parameter type is assignable from Message/KafkaNull before returning the raw input. If it isn't, construct a MessageConversionException describing the failed conversion and call messageConverterHelper.shouldFailIfCantConvert(message, thatException) — passing the actual exception, not null (note: shouldFailIfCantConvert short-circuits to false whenever the throwable argument is null, both in the default implementation and in ours, so the framework needs to construct and pass a real exception here, not signal with a null throwable). If the helper returns true, throw the exception instead of returning the raw message. This reuses the exact mechanism #1168 introduced rather than adding a new one.

Current workaround

We've worked around this downstream, in our own Kafka error-handling layer, by pattern-matching the resulting ClassCastException's message (checking that the class being cast from is GenericMessage) and reclassifying it as a MessageConversionException after the fact. It works, but it's inherently fragile (string-matching an exception message), and only possible because we own that error-handling layer — most Spring Cloud Stream users wouldn't have that option.

Guida per i contributori

Apri la guida per i contributori

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Direzione di ricerca

Inizia in spring-cloud-function-context/src/main/java/org/springframework/cloud/function/context/catalog/SimpleFunctionRegistry.java, in FunctionInvocationWrapper#convertInputIfNecessary e nell'early return di KafkaNull. Confronta questo percorso con il flusso di MessageConverterHelper introdotto per #1168, quindi riproducilo con un consumer Kafka tipizzato che riceva un record null. Il lavoro è completato quando i payload KafkaNull incompatibili producono l'errore di conversione previsto, mentre i consumer compatibili con Message/KafkaNull mantengono il comportamento attuale.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Valutazione

Stack tecnologico
java, kafka, spring
Ambito
backend
Tipo di issue
Bug
Difficoltà
4/5
Tempo stimato
3-5 giorni
Stato di attività
Attiva
Chiarezza
Specificata chiaramente
Idoneità per principianti
55/100

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.